Method and configuration for improving the dynamic range of an adaptive recursive network for processing discrete-time signals

Abstract

In a method and configuration for improving the dynamic range of an adaptive recursive network for processing discrete-time signals, gradient signals are formed from output signals of the network with respect to network parameters. The output signals of the network are compared with set-point signals to form differential signals. Signals of a signum function of the gradient signals are formed. The signals of the signum function are multiplied with the differential signals to form correction signals proportional to the product of the signals of the signum function and the differential signals.

We claim: 
     
       1. Method for influencing the dynamic range of an adaptive recursive network for processing discrete-time signals, which comprises forming gradient signals from output signals generated by the network with respect to network parameters, providing set-point signals, comparing the output signals generated by the network with the set-point signals to form differential signals, forming signals of a signum function of the gradient signals, and multiplying and signals of the signum function with the differential signals to form correction signals proportional to the product of the signals of the signum function and the differential signals. 
     
     
       2. Method according to claim 1, which comprises forming a proportionality factor for the correction signal from the reciprocal value of twice the number of the parameters of the network to be adapted. 
     
     
       3. Configuration for influencing the dynamic range of an adaptive recursive network for processing discrete-time signals, comprising means for forming gradient signals from output signals generated by the adaptive recursive network with respect to network parameters, means for conducting set-point signals, means for comparing the output signals generated by the adaptive recursive network with the set-point signals to form differential signals, means for forming signals of a signum function of the gradient signals having an input for receiving the gradient signals and an output, and a multiplier having an input side connected to the output of said signum function signal forming means and to said comparing means for multiplying the signals of the signum function with the differential signals to form correction signals proportion al to the product of the signals of the signum function and the differential signals. 
     
     
       4. Configuration according to claim 3, wherein said multiplier has an output, including an amplifier connected to said output of said multiplier, said amplifier having an amplification factor equivalent to the reciprocal value of twice the number of the parameters of the network to be adapted. 
     
     
       5. Configuration according to claim 3, wherein said means for forming gradient signals are in the form of a gradient network forming the gradient signals, said gradient network including a recursive subnetwork portion and a non-recursive subnetwork portion emitting the gradient signals. 
     
     
       6. Configurations according to claim 4, wherein said means for forming gradient signals are in the form of a gradient network forming the gradient signals, said gradient network including a recursive subnetwork portion and a non-recursive subnetwork portion emitting the gradient signals.